A practical way to use clustering and context knowledge for software project planning

The use of empirical data from past projects for project planning is gaining increasing importance in engineering-style software development. Since software development projects are unique, experience from these projects cannot be reused directly. Nevertheless, common patterns can often be found when comparing past projects. This information can then be used to better support project planning. The article sketches the SPRINT I technique for project planning and controlling. The approach is grounded on the usage and analysis of context-oriented cluster curves. The article focuses on two aspects: How to identify similar projects and build so-called clusters with typical data curves (such as effort distribution); and how to characterize these clusters by using context knowledge. This allows for assigning a new project to a cluster in order to obtain a prediction. Results from an evaluation with data from 25 projects show that the technique provides a practical way to increase the accuracy of software project planning by using empirical data.